Subscribe
Share
Share
Embed
RiskCast documents what happens when you make AI agents structurally indispensable — not optional. Host Stefan Friend is building OpenClaw, a self-hosted multi-agent system that runs his businesses, and sharing the real experience: the good, the bad, the ugly.
Each episode explores the practical reality of living with AI infrastructure through conversations with builders doing the same.
I've got aI've got a friend that's kind of like a surrogate younger brother that's in film school up in Chicago.And he's interned with one of the racing teams up in Mooresville helping edit content.I asked him if he'd help actually edit.So he'll be the one kind of helping take like we've, they'll give us the video back and he'll help turn it into whatever this looks like.Yeah, this is awesome.I'm a little bit curious before we jump into AI stuff.Like, when did you start at Tabrace?What brought you here?And you have quite a bio, like you're working in Bank of America still.You're dabbling in some Alfred and AI stuff, and you are running this co-working space.How does that all come together?Yeah.Well, you probably remember the original face of Tabris with the much longer beard than me, right?So I was a part of some of the original conversations about a year and a half before that when I think there were 7 or 8 of us originally thinking through the idea of a co-working space, but tech-themed.You know, Advent and hygge, of course, I think have become staples for coworking in the city and Packard was the closest thing I think before this to a more of a tech focused.smaller, in Uptown made it a little bit more challenging, or that's what we thought we were trying to find a space.I was working with a lot of that same group, helping with product for 2 different software companies.And so I was not as actively evolved initially.One of those companies was fintech focused, and we were growing pretty significantly late to 20 19, right around the time Tabris opened.And then heading into early 20 20, starting to try to put together a real investor data room to go after a series A was the intent.And then March 13th is kind of seared in in my brain is when the world for us really started to fall apart.And I think a lot of other people.fast forward 4 or 5 months later and there was really nobody else left to run tabris and the fintech had kind of fallen apart so jay for asked if i'd step in and i think at the time it was as much keep the building from burning down and just kind ofkeep a pulse on everything.But if I'm here and there are other people kind of hanging out, I might as well try to figure out things.And so I became the kind of a final partner left and managing partner day to day.And I've been doing that ever since.Very cool.A lot different having a physical space than software though.And I, I like building, I like experimenting.So I've always continued to do that.So I've, since I've been running Tabris, I've all, I've done a full stack Java bootcamp.I've had 2 different product roles for Series B software companies, one based in Silicon Valley, one based here.And then I got recruited to lead the commercial card mobile app for Bank of America.And that was the most interesting thing that one of the banks has asked me to do.So I thought I'd take a shot at it.I'd also never been in a company bigger than two hundred people before.So it seemed like it was worthBranching out, having a different experience.I've advised fintechs for years.And the biggest gap in doing that had always been governance and regulatory navigation or understanding those processes.So experiencing it firsthand felt like it was worthwhile.It sounds like your expertise and like governance, the stuff is super handy for building stuff with AI.I think it's helpful.And as hopefully we'll get into a little bit and what you, we shared, and I think some of the themes from everybody at the community night, I tried very much to be thinking about that upfront and you'll get some of it right.You'll get some of it wrong and hopefully you keep dialing it in andlooking at what everybody else is doing to learn too.Yeah.A hundred percent.Well, thank you for sharing.Yeah.I was thinking similarly, love to better understand.We've had some conversations off and on at different events andI know that you've been pretty integrated into the tech ecosystem in Charlotte, and especially the health tech ecosystem for at least a decade at this point.I'm sure it feels long and short, but love to hear more about you and kind of how you got here too.So when I moved to Charlotte, I didn't have a network.I didn't have a job.I didn't have anything.So it's like the coworking space became a gateway for me.I joined Advent Coworking.And at that time, it was not the corporate space like it's becoming now, I think, with Hooga being acquired and acquired.But it was like one founder or initially it was 2 founders running this community.And they happened to run it out of a building.And that was an awesome experience for me and an ability to meet people, to get mentorship.that's something that like I'm coming from a culture where mentorship is not that common, I guess.Maybe I was like not the one exposed to it.Whereas here in the US, especially in the coworking space, it's like, how do I write this email?Can I ask this guy sitting next to me?And you start conversations, you start understanding people and people actually help.Like Juan would spend, and Juan was big even ten years ago, he would spend like 30 minutes talking to me.Like Igor, this is not the right approach.And he let me pitch my crazy idea at pitch breakfast, even though it didn't fit like the format.Yeah, he just did it.And I was like, this is not like Belarus.And I loved it.I loved how open Charlotte was, how easy it was to plug in.So to cut, the story can go on forever.So the coworking unlocked 2 things for me.One thing was the mentorship lane.There was this emergent organization that was not yet called Innovate Charlotte, but what became Innovate Charlotte.And I joined that organization and then we went to MIT to learn the model of mentorship and I was hooked.Before I tried playing on my own and I had like a community MBA project with the founder of the coworking, Kevin Girunas.It was a fun project, but the mentorship piece failed miserably.And then going to MIT and seeing, oh, what these guys have figured out over the last 20 years was a mind-blowing experience.And then I got an opportunity to run the mentorship program for about 4 years at Innovate Charlotte.And in parallel, in the same co-working space, I met a guy, Brian Formato.I did a workshop on negotiations.That was my thing back then.And he attended my workshop and he was doing some work with a crew partners, a company in town.And he said, Igor, you are doing negotiations.I like your workshop.I'm doing something on conflict resolution with a crew partners.Would you like to partner?and do the workshop together.It blew my mind.And we started working together and we've been working since.So now I kind of have 2 lanes.One is leadership development lane, still working with Brian Formato in group management.And it was quite an apprenticeship.And the other piece is mentorship.In health tech space, it's advisory 5, like doing, supporting local health tech startups.But I'm also passionate about the model of mentorship overall.So I am right now a visiting innovator at Northeastern in Boston, helping them activate their mentorship community and working with different communities that are building or running mentorship all over the world.That's my passion.I love it.And definitely seems like the 2 lanes have a lot of crossover too.So it makes a lot of sense that you can be working on a project with Northeastern and learn things from what you're sharing, but what you're getting from the people you're interacting with and then bring it into the other areas you're working on too.A hundred percent.Yeah.That's really exciting.And so you are hands on with lots of different teams, lots of different projects across these 2 lanes.When did you start really getting interested in the growing AI movement?We'll call it that for now.I think I was curious about technology.Back in Belarus, I had a digital marketing agency.I was not like building stuff, but I was playing with stuff.And even as a kid, uh,In Soviet Union, my parents got me this old Soviet machine.It didn't have windows.It had like, what did it have?It had a Soviet version of DOS operating system.It had a bunch of books.I think some factory sold it.Books on like C, not plus plus, just C, Pascal, Fortran, like a bunch of books.I didn't have any games.I needed to make a game or something.And that was kind of my childhood experience with technology.Later, having this marketing company, working with engineers.My partner was an engineer.I didn't go deeply into like actually doing, but I loved thinking and kind of translating ideas into specifications that could be implemented.That was always fascinating for me, how you can make that happen.But specifically AI stuff, I have a friend actually from high school in Belarus who happens to be a manager at Meta.And we met a couple of years back.And since then, we've been having standing calls every 6 weeks.And he's been my source of inspiration, all things AI.And I think he was the one who told me, we're looking to ChatGPT.At that time, it was not yet a chatbot.It was just a developer console where you could play with ChatGPT-free.Yeah.I played with it.I liked it, but I didn't love it.Okay.So it felt kind of.So this is early, midtweenty23?2022.Like the spring of 2022.Yeah, because Chat came out in late November of 2022.That's right.So maybe 6 months before that.So I played with it.It was fun.I was curious about it.It was not exactly there, but it was on my radar.As soon as they released, I think, ChatGPT4, wow, I wasOkay.What about yourself?What was your kind of Yeah, I was on the wait list to get onto ChatGPTWithin hours of it being launched, I had signed up for the waitlist and then I don't know how they were tranching people getting access but late November, I, I had access, and I was playing with it and I'm like, this, this feels like a big step from.what people have been calling AI up to this point.Even the relatively early versions of the chat bots felt like they were different than needing to haveA clearly defined sequence of things or workflows and using AI when it really was you were automating something that took manual work before.This felt like you could actually engage with it differently.And so I've been, I'd say, a fairly heavily user of all of them since.What was your initial kind of usage?What did it look like when you started?Yeah, I was still at Mighty Networks at the time.And I think one of the first things, probably similar to Andrew that we were learning from at the OpenClaw Community Night, I was trying to stay more in a product management, product development lane.So if I was having a conversation aroundYeah, customer discovery, trying to capture notes and then share those notes into a chat, go back and forth with it to kind of refine and maybe even challenge some of my product thinking so that way I could try to move forward on.Okay, here are the maybe the questions I want to go back to the customer with, or, you know, I want to go internally to a tech team, a marketing team, leadership.Yeah, what do I need to take the next action on to move this forward?I think that was one use case.The other was innately at a startup in particular, but seeing even at Bank of America, product manager still ends up with a very broad range of responsibilities and typically covers some amount of gear work.It's necessary, I think, both for the traceability of work being done andAnd giving everybody on the team that's working on building clear point of reference for what we're building, why we're building, and what those requirements look like when you're done.But it's tedious.It's monotonous.And I think especially different versions of the project tools, whether it's AHA, it's JIRA, I've used Redmine, an open source version of those.they look just a little bit different, that every team's version can feel a little bit different.I'm curious, for Jira, it's more like manual updates, so you need, like, the information.It was still manual.How did you do it?Like, how did early versions, like ChatGPT4, how did it help you?Yeah, so I've typically had a template of that PRD, the product requirements document.And so that's I like just a running Word doc that starts to get really long, but you can share that into a chat and then have that conversation back and forth to pull out, okay, here'sHere's the acceptance criteria.And it maybe started initially of just kind of that of, okay, I gave you a doc.I want acceptance criteria.Let's start with what do you just give me?And it's aggregating toward the means, so it gives you something.You go, okay, that format doesn'tIt doesn't quite work for my team.Here's what we'd like.And you start tweaking that.And now you start to fine-tune, okay, if I upload X and I want Z out, here's the prompt Y I need to be able to give in.And you customize a prompt.And so now I've got a prompt that I can use repeatedly togiven that i have the information to provide and i get something more consistent out and i think for me it's felt like fast forward several years that process has keep getting better and it's felt like in the last few months now i'm close to reaching the point where you don't necessarily need to be providing thatWhy in the middle, if you've outlined here's this process up front, you could create a project or upload markdown documents and kind of outline everything.And now we can have a conversation.And then at the end of it, say, okay, help me outline this project.And it might be able to build the product requirement documents, including all of theepics with acceptance criteria and you're in a completely different lane as far as being able to start working on something if not handed directly over to cloud code and have it start building it.I love it and I love how your initial use case is so sophisticated and is already matching the current workflows.Mine was more like, okay, let's talk about the meaning of life.Let's help with a couple of emails.That was absolutely happening.And frankly, just with Tabris, it was probably a lot closer to more ad hoc one-offs that maybe we were trying to fine-tune an event proposal.Those were probably the things that we did the most with the tools trying to figure out how we'd get better.Because II'm not an event producer.I'm not an event manager.I never have done those.I worked in country clubs in college to help pay my way.And so I'd be the bartender at an event and I'd kind of see how people would run it.But being the person trying tosell events, make sure the hosts have everything to plan and execute them was very foreign to me.And we've got a massive, massive to me event space that is pretty cool for people to be able to use.I didn't know what I was doing.I liked having an AI overlord to kind of tell me what, what should this look like?this idea of AI overload resonates.I think for me, like the moment where the next shift happened beyond just immediate support was when I discovered Claude.And at that time, ChatGPT didn't have projects or maybe they didn't have projects in the, I think the web version was always ahead of the app and I was using the web and then I didn't see any projects and I tried Claude.Oh, it has projects.Yeah.And I think it was relatively recently, maybe end of 2024 when I noticed that.And that's where it clicked that I can use AI to help me around my life, not just to answer my questions.Yeah.I started using Chachapiti most heavily.And then I think Gemini probably next because I'm heavy Google workspace users.And so they sometime, maybe it was 2024 or early last year, they unlocked some of the AI features for the business workspaces.And I had a Claude account.I had a Grok account, DeepSeek.Like I'd test them.Didn't really use them very much.I liked ChatGPT's projects until at some point that I realized they didn't really maintain context very well.It was more of an organization tool than a functional tool to me.Or, which was not how I thought it was I kept trying to have conversations, assuming it knew everything in this project and that I'd end up with it hallucinating, or only providing some context and that was frustrating relative to, I think the relatively recentcolloid projects that are significantly better at maintaining context or being able to go search within all of the previous conversations.That's felt like a complete game changer to me.That makes sense.So when what was the moment when you felt like it's the next shift?You told me about this product requirements document workflow, how it shifted.Was there any other moment where you felt, wow, I'm a different person now or I live differently or operate differently?Yeah, for me, the last model or the in in the end of 2025 that Claude released and you started to.So just a couple months ago, just a few months ago, started to really hear, I think, increased noise around what it could do for building.Mm hmm.And I think Bart had shared at the one of the more recent talks about AI or sorry, the agentic AI is still a relatively new term within the last year, roughly.Last year, I followed along, but I didn't really play much with it.I just, I was skeptical enough.I didn't want to spend time trying to figure that out.And I still wasn't doing that late.I was just following along with those, but the, the ability for it to start building with Claude code seemed like it was shifting.And it didn't feel like as much noise as much signal of what had just changed.And so I started playing with it a little bit.And then I sat in on, uh, Charles first one-shot labs cohort, 6, 7 of us trying to take one document of context.However, it was that you created that his suggestion was let's have a conversation about a problem, create documentation from that conversation and then hand it into the tools and let it build.And it did a pretty decent job.I took my existing product requirements document template, engineering spec template, and then I created a new artifact that I called the design contract template.Hmm.Because I felt like, to me, the missing piece was, okay, I can give it these 2 things and it will build, but how do I guide it toward building in a way that I think is going to be simplistic for the user?And so I essentially ranted at Claude for 5 minutes ofI love Duolingo.I love Acorns.I think that they have simple user experiences.I love how they work on web and mobile apps.I'd like to bring in the best practices from these.And there were a few others as well, but use that to create the design contract.So I used all 3 of those templates with that conversation that I'd had with a friend in order to create the documentation to hand Claude code and see what it came up with.And to me, even 3, 4 months later, you still are going to end up having some back and forths.But what it was able to build in a day, even with those back and forths and that human in the loop, I thought was so cool.What did you build?A friend and I had a conversation around paving construction company business that he essentially is the technology person for.And that they have challenges with the existing construction project management tools.They are not perfect fits.They maybe get 80 to 90 percent of the way done.But that last 20 to ten percent creates enough headaches that it's frustrating.It takes a disproportionate amount of time to work with.So we started building an MVP version of that.AndIt was proof of concept enough that he's kept working on it.And I think is actually trying to refine it to take to his client.And maybe it becomes something that becomes a real tool that could be brought to other painting companies.Very cool.So it's on the path to become a real project.Yeah, we'll see.I didn't keep building that.I kept building some other things.Back to the event stuff is, I think, the biggest gap at Tabris.We created an event proposal generator.We created a tool to better track the single event P&Ls.Some things to try to make it a little bit more consistent instead of things living in different docs or workbooks in Google Drive.I love it.So I'mObviously, very excited about all this.I'm playing heavy user, but I want to learn more about kind of those same things.You've been playing with these tools before me, even when they were notthe consumer facing chat bots you know what has the last few years of you using the tools with and then then we'll start to transition into like the the agentic ones uh date really start to dig in thereSo I think for me, given my 2 lanes, which are very human focused, most of the focus of AI was less about building stuff for these 2 years for others.It was more helping myself, like taking my journal and figuring out what are my blind spots, what I'm not seeing.And I was blown away.By how effective Claude was able to identify things I needed to kind of refine.So I mentioned Charlotte gave me mentorship.I thinkI also go to a philosophy group, which gives me better thinking.And honestly, my education in Belarus, like studying Marxism, Leninism, whatever, was so remote from philosophy.So it was like a blind spot in my education.And I thought that philosophy, it's like nonsense or like crazy people are talking, we have science, who needs philosophy?And then over the last year, kind of ingraining myself in a community of passionate people, I learned, wow, the best thinkers are actually philosophers.And we as society are lagging behind.What we are thinking about the world is actually what they were thinking 50 years ago.Maybe some are still thinking, but they're way ahead.That was like blowing my mind.So if I need to refine my thinking, this is like philosophy.And then AI, this is leverage.So this is kind of I'm thinking about this.How do I grow professionally and personally at scale, even as an individual?I love that.To me, it feels like those are naturally complementary pieces in the direction AI is going in that as AI continues to get more autonomous, moreWe'll use smarter, you know, heading toward the AGI or, you know, the super intelligence.The critical part to me is what is the human complement to this?And if we aren't expanding how we think and think through different perspectives and the ways that AI continues toreally help us have a positive impact on the world, the alternative is AI just runs away.It quite may run away.It's still possible.Yeah.The cat's out of the bag.There's no getting it back in.Non0 chance.But to answer your previous question, so I didn't build any products.I was focused on kind of building my routines, building my thinking models.Figuring out how do I approach my coaching work?How do I approach working with mentors?How do I make this as impactful as it could be?So that was my focus.But I think in the fall of 20 25, this agentic AI was rage.I decided to take a course on the agentic AI.Oh, cool.It taught me, oh, this cursor, there are all IDEs you can use.You can set things up.It was still kind of manual.i don't think there was cloud code in it but it gave you like examples templates and i said oh that's much better than i when i was trying to learn like python 3 years ago so much more kind of actionable and then by some chance oh claude has this cloud code i connected this and this is where it kind of blew my mind it was maybe in december that i could actuallyNot just kind of go line by line and still like figure out from pieces of code and whatever I'm building, it's still primitive.I can build something real.What I took, I took away my website, Advisory 5 from Squarespace, which was, I didn't like it.And I just said, let me build this with cloud code.Yeah.And that was kind of one of the actual first projects I decided to do.It wasn't a great website initially.It definitely had a lot of security flaws.It had a lot of mobile interface flaws and other pieces.But the fact that it was running and it was better than the previous website, I think something shifted in myI think you hit on something that's definitely started to change a lot in the last few months to me.This is One second.I feel like that dropped more than it should have.This is why we edit.Or maybe we'll leave it.I don't know yet.You have to be willing, I think, still a little bit, even a few months later, but definitely late- to maybe be a little bit uncomfortable going into the spaces that are historically reserved for the engineers.I don't think you had to be completely technical, but you had to do something that was not just have a conversation with the actual engineers and let them just go do the stuff.Yes.Yeah.What have you been doing in the last few months?Because that's starting to hit on where you really started to play out with the agentic AI tools.And then there's been an acceleration in what they're capable of and what people are doing with them.And we started to see the other night you're doing some cool stuff.So for me, Cloud Code was an unlock.So I had a number of small projects helping me.Like I didn't like Doodle for scheduling.I built my own tool.Then my insight, oh, I can use my tool to kind ofhow do those tools connect?So I was thinking, but it was like individual projects run by Cloud Code.You figure out, oh, there needs to be a backend.There needs to be a database.For me, it's like asking Cloud, what are the options?Okay, Verso, Superbase, Railway.So you kind of figure out as you go and set them up.But I think the aha moment was when I heard about CloudBot or OpenClaw.Yeah.And all the rage, people were talking about personal assistance, personal efficiency.That didn't resonate with me at the time.You'd already done a lot of that.I was doing it organically, and I thought, okay, yeah, you talk to Telegram versus the regular app.Okay, so what's the rage?And then I heard the guy who built it talk about his use case.and his use case was he was crazy about coding and at that time cloud code couldn't like automatically build something like long over a long time every like 30 seconds it asks you to confirm something yeah it's fine but it's painful and then no i can build from my phone i need to get the mac me that was a shift and i was surprised my biggest surprise wasi watched a lot of videos talking about open claw cloudbot 99 percent of those videos were about just personal efficiency security no one was talking about using this as a tool to build only the guy who actually built the tool that was his main use case and this issomething that resonated with me.I got the Mac Mini.I immediately integrated all my projects and gave it to Claudia, which is my OpenClaw agent.She is responsible for all those pieces together.This is where really things started to shift because you don't need to spend sleepless night clicking on Cloud Code and thinking.You can build the spec, give it to Claudia,she works on it and yeah i don't have to be in front of the computer to do the work so that's been exciting just to name things that i've kind of worked on uh one of the things just to help operationalize my life is thane it's a crm system i built for myself but i decided why not offer it to people who are working like mostly by yourself like a fractional professionalI made an announcement on LinkedIn.A few people said, I want to be a volunteer or better user.So I have a few of those.And that really changed my thinking because people give much better ideas than I can come up with AI.people are underrated, a hundred percent.So that's one space.And the use case is like I have all my contacts.I have my calendar.I have tasks.I have projects in one place.So it's my kind of Margaret is saying thing to keep you sane.Yeah.So that's the main life OS.LifeOS and then if I need to send like 50 emails to different people in different groups, Claudia already knows those people.She can tap the CRM.So I figured that I need to connect those different tools I'm building with APIs or somehow so I can use my slot pick, which is a tool like Doodle replacement to create links for scheduling.It has my Gmail.I think it has contacts.So it's kind of emerging as a bit of an ecosystem that I'm trying to still put together of, again, how to build leverage for what I'm doing.Yeah, I like that a lot.And it feels in some ways to me like you took a different approach, but trying to solve some similar problems to how I did in, I thinkYou have an idea of the system that you wanted to be building toward and how you could use open claw to really better understand you, your life, your projects, and create a more cohesive way that they're all connected and there's less of the jumping back and forth finding different things.So tell me a little bit about Alfred.first of all, why Alfred?And how did you come up with the idea that you need this type of an executive assistant?Yeah.Well, the first part, I'm completely a nerd in case the comic book sleeve isn't giving that away.So I thought that I wanted to have my AI agent be a callback to some of my favorite characters.It didn't feel like it made sense for this tool to be Jarvis.I don't think it's quite there yet, 1.2, I'm just more of a Batman fan.SoI view Alfred's character in the Batman lore as kind of like the pertinent to Batman, to Bruce Wayne.He is essential.Batman does not run, doesn't function, doesn't exist without Alfred.And I think in some ways I'm hoping that this tool continues to evolve in a way that it's helping me in that way.Am I capable of living and running without Alfred?I certainly hope so.I've made it this far.But it's so much more powerful and capable already and it's going to continue to get better.I'd love to keep building toward the type of cohesive ecosystem that I think you've, to me, it feels like are a couple of steps ahead in lots of ways, where Alfred's got so much of this context, it evolves toward, instead of me having to initiate, how does it become a little bit more proactive?And I think that that's the direction it will go.But soat its core, that feels like it's essentially an executive assistant.The best executive assistants that I've interacted with, and not ones that I've had, but a couple of my mentors have had them, they are so plugged in into their lives, into everything that it takes to run their business and personal lives, they're the oil in the machine making sure that everything is running.And I think that there's a way that this keeps evolving toward that and is connected and plugged in across personal and business.I have a question on this because I love Batman out of superheroes because that's the only superhero whose superpower is actually technology.yeah so essentially alfred is your superpower on the one hand so you are the center on the other hand you are saying that uh you would like alfred to be more proactive yeah i think eventually so but this is like how like even philosophically i'm wondering if it becomes more proactiveWhat's the superpower angle then?Do you maintain the superpower angle or does it become some new category of support?Yeah.Well, so I guess walk through a little bit more of one of the primary use cases at the moment whereIt's not completely proactive, but because there's a rhythm at the moment, it's helping our team at Tabris, at least, do things that would take a little bit more manual action previously.So I keep coming back to events because that's, I think, thearea that we have the most idea of what that workflow and pipeline should look like, but there's just a lot of nuance.Working with an event inquiry comes in, somebody wants to host an event, maybe it's a workshop, professional, maybe it's more personal like a birthday party or a baby shower.There's been some nuance there in trying to manage that.So that's one of the areas where Alfred got set up initially to monitor the generic inbox.It's got a cron job polling the inbox on about a thirty minute cadence to just check.We don't need real time updates.We need something consistent, though, and then be able to surface those inquiries with key details into our slack so we can take action on.That was where it started.as I was trying to fine tune, I kind of built the system initially of Alfred's the, the one that we're interacting with, but I built a monitor agent that that's the one that's really pulling everything.And then a drafting agent, the drafting agent was the one that initially felt like it was struggling with tweaking of, uh, I was getting drafts to respond to spectrum business.It's like, okay,Monitor surfing something that Alfred's saying is important and handing off the draft that's not.We need to fine tune this connectivity a little bit so it's laser focused on the actual leverage points.And we've got that dialed in.So it's helping us surface.Here's an actual inquiry.Here's a suggested reply.Then we still are the ones manually sending it at the moment.I think we want to get to the point where we have enough confidence.that we aren't the ones sending it, and the initial back and forth is Alfred.Then is it Alfred or is it more of a Robin kind of character?Yeah.Yeah.It's starting to be more of a sidekick.So you're closer to Robin.We'll still be calling it Alfred for now.Maybe we spin that out and it becomes more of its own true dedicated agent.Yeah.I love that idea, and I have a question that I'm figuring out, honestly, on my own.I'm diving.One of my use cases for AI is to help me learn faster.And what I love, I didn't have this as a college kid.I didn't have college textbooks back in Belarus.Okay.Our textbooks sucked or we didn't have them or we had like old Soviet ones or new ones that were not good.So when I was exposed to college books, for me, it was like aha moment.And now this college book is the concept I use to learn anything new, like even philosophy or cybersecurity.I'm pulling different materials.I have books.There are some great books that are available in Russian without like DRM protection.So I can feed them to a folder to pull out co-work.So I put this all together and my prompt is build me a college level textbook one on one.Cool.For my use case.yeah include examples include multiple choice questions include prompts and include answer key this is my awesome life hack so right now i'm working on cyber security piece and figuring out how like you build secure system what i love the concept of cia uh triad in cyber security which is like confidentiality uh what was the second one iaccessibility and what does the I stand for?Is it information?I'm blanking on the exact word.Let me look it up.Integrity.So I'll say it again.So CIA confidentiality, keeping information secret from unauthorized access.Integrity, ensuring information is accurate and not tempered with.And availability, make sure data is accessible when needed.I read a couple of chapters.I create a spec for Claudia to implement across all my projects.Oh, Igor, I found like ten vulnerabilities.We need to rework that.So I'm curious in your more agentic workflow that you are describing, how you approach security?Yeah.Or the whole CIA piece.I probably should maybe take a little bit more time to dedicate to understanding some of the security vulnerabilities myself instead of more innately trusting that I can send a prompt into an agent and it's going to be accurate in one, auditing, and then 2, fixing.I've taken that approach essentially.So II have typically started with cloud code prompt to audit the existing open cloud system and return back the security gaps that you think exist and propose a plan to fix it.I've then, the only thing that I'm using ChatGPT's codecs for at the moment is that exact same prompt.So then I ask it that prompt, they're both on the Mac mini, so they have file access in order to see the system.And then I compare and contrast the tool tools, what they returned, and then I hand it back to Claude Code.So, so far, most of my changes to my OpenClaw system have been built by Claude Code, me more directly interfacing as opposed to asking Alfred to work on himself.I just felt like it's been a smoother process.And I have the tool.I'd rather do it more directly than ask Alfred to do it, has been my approach so far.But I've liked that of being able to let these 2 very powerful systems work.run through the same prompt, create a report, and figure out if there's any gaps.And for the most part, the delta between their approaches has been pretty small.So it's felt like there's some level of confidence in what we're going to be working on to enhance.about once a week and this is a chrome job that alfred runs to initially surface it it it has a prompt where it's running through at anything in the reddit forums particularly the the subreddit for openclawand other RSS feeds to find where people have made comments on security vulnerabilities and enhancements.And it uses that to feed back into that prompt when I run it.I love that it looks like we have 2 different approaches to architecture.Your approach, and by the way, this is a great way to, when I make changes to Claudia, I also use Cloud Code.I don't ask Claudia to do it.So that resonates.So you're using Alfred essentially, if I got it correctly, to run this event management system.Yes.My approach is a little bit different in the sense that if I have a use case, I'm building a more deterministic system, not AI system, like an app for this particular use case.And then Claudia helps me support the app, but it's not Claudia, if it makes sense.Yes.In that sense, she's helping play the partner in building, but then it's It's on its own.It's on its own.It might have its own agent.It might do its own kind of automated stuff, but it's a separate piece versus Alfred is the kind of the thinker and the doer, the third partner.I love that there are so many ways to play with it.In terms of like confidence, you mentioned you don't have enough confidence yet for it to automatically send quotes.So what would need to be true for you to get this confidence?probably 2 things, and I think one is the human component first, and then let's test and experiment in partnership with that AI to see if it does it.Right now, and going back for years,There's felt like enough nuance in the back and forth with a host that's inquiring about an event to fine tune what their event should look like and that event proposal should look like.And that we feel like we've wanted to be directly human to human, even in the last few months as we've gotten closer to an AI that feels like it's got more empathy, more understanding of the nuance.I don't feel like we're quite there yet.And that human part is, okay, even for us, sometimes we're not sure what is the best way to understand what their event is trying to be and how we help them.And maybe if we were better event planners or we run other event businesses, we would feel better and we'd be able to better design and integrate with Alfred and how we run that.But I think we'reWe're closer than we were 3, 4 years ago when we started actually having paid events.But we're maybe not quite there yet.I'm curious because I am facing kind of the same dilemma in a different space.So I mentioned like health tech and mentorship is one lane, but they also have this corporate lane of leadership development.And one of the things we do in that space when there is like a coaching engagement, there is a three sixty degree feedback.yep yeah i've built a tool okay for that it's called alarum i'm using like shakespearean uh words that are available for domains uh to play with that and this is my tension point like i got an okay from my boss at groove management from brian igor if you want to use your tool you canBut now it's like pressure.We have the tool that we've been using at Groove, like a professional but very old-fashioned tool to run those things.Now I have a coaching client that I can use this tool.And yesterday I was struggling.Okay.What did the more antiquated kind of traditional 360 feedback tools look like?And when you built your tool, what was the initial difference you were trying to create?I think the initial, honestly, it's just old and expensive.So you pay a lot of money for super basic functionality that doesn't have any AI assessment or like analysis built in.It gives you almost like raw data.So would this be not dissimilar from some of the corporate tools like DISC or SDI or, you know, Myers-Briggs type of thing that you you kind of you have a leadership person that's wanting help.They do this assessment to try to get a baseline.They get back a report and then you're the one that helps them understand where they are versus where they want to go.Yes.And we have those as well.And I want to go there as well with my tool.But right now, 360 is more like feedback around like, for example, if we were doing it for you, it's like if you have a boss or chair of the board or someone you report to.they provide feedback on your work, then your peers, people from different groups.so it's usually either a conversation and filling out the form or just people fill out the form like communication as stefan is precise and clear in his communication he is right now at 3 which is good we want him at 5 which is exceptionalAnd the person gets assessed in all those areas.And the challenge assessment-wise is to make sure that the form, if I send it out, or if the person I'm working with sends it out, first of all, that it works perfectly, that the information doesn't leak, it's secure.And then that I'm able to process this information the same way I would do it in a tool.It's not a very sophisticated use case, but because I'm like essentially using it for a real leader and a lot of people on his team will be involved in answering, like the stakes are higher.yeah well and i i think i see where there's some similarities to what we've had challenges with because of the human element like even forgetting the the security vulnerability potential which i think is is very important you're getting qualitative narrative input in these these fields in this formAnd, you know, how do you quantitatively use that?You've got your one through 5 metric, it sounds like, as one piece, but then you still need to do something with them writing a few sentences about the communications.If they want, they leave comments, and then those need to be de-identified.there are pieces there.Technologically-wise, it's not hard.And the problem that has been solved before, it's more like I'm sure Alfred could also do a great job communicating with clients about your events, and it would do a great job for the most part.But the bar from getting like to 95 percent confidence to 99 percent confidence.Yeah.I feel this is kind of for me, it's almost 90 percent of work, like 90 percent of work to get to this ninety and then 90 percent to close the gap that is remaining.Yeah.It's not on the technology.It's like on figuring out that the users I'm giving this to actually get the right experience and it doesn't blow up.I feel that it's like the inverse sides of the Pareto principle.Yes.Yeah.So it feels like something you're going to keep working on, though, and refine, right?I asked Claude yesterday, so what do you recommend?I'm struggling here.And it said, Igor, is it a hobby project?Or is it something real?If it's real, you should do it.The blast will be not so good, not so big if it blows off.You can always switch to your corporate tool, but you should do it because this way it becomes real.I like that.Yeah.Ship it.See what it breaks.And either it's not a big deal and you tweak it or you fall back.For me, it was more like, do I spend another 20 hours on this?Because it's done.But I and honestly, I spent half a day yesterday night and I'll be working on it like today to get it to the point where I get this confidence where I'm able to use it.And you said you asked Claude or Claudia?I asked Claude.And this is, by the way, I, Claudia, and you noted it clearly.Claudia, I don't talk to Claudia about this stuff.Claudia, for me, is my building partner.Yeah, she's my engineering team.I use co-work to build my textbooks or to kind of help me learn deep things.And I use just the project where I asked Claude to act as an executive coach for me.So then it pushes back.I'm an executive coach, so I know exactly what I need.It pushes back, it asks questions, it identifies patterns.And this is where I have those kind of discussions.That's cool.As we've had some of this conversation, and certainly on this topic, I'm feeling like II need to refine, I guess maybe the cloud code instructions to help it do that with me a little bit better I. Some of it's probably some of my neuro divergence profile but also I think it's just the.breadth of potential opportunities in the world today that it's very easy to get distracted with a hobby that could potentially be a project.And how much time do you spend on something when the higher leverage point is something else?And sometimes it's challenging because it's also just fun.It's easy to get lost in it.It is easy to get carried away.And that's one of my personal things I'm working on.Like,identifying things that need to get to completion yeah versus trying to like build for the sake of fun I like it umI want to hit on a couple of things still that we haven't gotten to yet.And then realize sometime in this conversation that we didn't introduce this brand new podcast.So we'll take that.We'll just do this at the end.And maybe we fix that in post or maybe we end on it.I'm not sure.We're going to figure out some of that as we go.But whether it'sjust the AI tools as a whole or specifically your open call system, what do you think are the biggest mistakes that you've made and then what did you learn from them and how did you fix it?I feel lucky that I haven't made like super huge mistakes, even though initially it's like you get carried away and you get a little bit reckless.One small mistake I made was thinking that like I can build this as a prototype and use like prototype stack, like Python, fast API, whatever is like cloud recommended, I would be building.And I realized it doesn't make sense.The speed of, like, what was, like, yes, before, if you were prototyping, this was the stack to use.Yeah.And this is in the training data.But now you can use, like, for prototyping, you can use the actual app stack, and it will be the same speed of delivery.yeah so i have a project that was built on this fast api in python that i will have to redo okay to do actually like production ready i like it but it's just not the stack i want to keep it's not as robust so that's mistake number one mistake number 2It's not necessarily a mistake, but thinking around the whole architecture from a security standpoint, from a user standpoint, from the very beginning.So I'm doing version 2 of my mentorship platform.And I realized the platform I have now, like TractionFive, it's great, but it's 2020 SaaS standard in everything, which is both good and bad.So instead of jumping in and building right away, like I did with other projects I mentioned, I'm being very intentional in figuring out what are the use cases, what are the test cases, reading the text booked on security, because there I'm going to actual clients.One client in Iceland told me, Igor, if my data leaks, I'll come and cut your throat.Yeah, the stakes are real with that, right?So I'm being very intentional.So with some projects, you don't have to do that, but with some projects, you do.And honestly, one more kind of not necessarily mistake, but thing I should have done, I played with Cloud Code, then I switched to ChatGPT and didn't have one project that kind of knows me for some time.I had it, and then I didn't have it.Now I have it.this time that I didn't have it I feel I lost clarity quite a bit so this is my takeaway I need like a thought partner even if it's not like Claudia but thought partner that knows where I'm at how everything is connected and that pushes back on my ideasI like that.Yeah.And they're better than ever at doing that exact thing if you're intentional about getting it set up and making sure that it understands who you are, as much context as you want to give it, and all of the projects that you're working on and everything that's important to you as well and how you look at things now and building things.What about yourself?What's your lesson learned or lessons learned?I think similarly, the security vein is really important.I thought I was being intentional up front, but I evidently could have been a little bit more careful in actually flipping the switch on a couple of the communications channels, maybe the prompt and connecting them.One of the channels that Alfred and the monitor agent are reviewing is my direct WhatsApp with my wife.So that way I can try to do a better job of keeping track of the important things in our life, managing kids, managing events, maybe it's grocery list.And you set up that direct WhatsApp connectivity to,And it was supposed to be only replying to me.But multiple times in the first 2, 3 days getting it set up, it was replying things to my wife, but not answering like a question.It was just sending like a long reply.Sometimes it was an errorSo it was sending things that it should have been notifying me about as far as just an alert, but to her.And so there were some wires crossed there that I thought would have been more clear up front that it should not be doing that.And I think there was just a little too innate trust in the system as a whole, how it was designed, that it was connecting things properly versus being moredirect in how I was setting up that connection and explicit about read only, do not reply under any circumstances, turning it off.And just to clarify, so you had, it's WhatsApp, correct?Correct.And you had your number and your wife's number?Correct.And Alfred had access to both?So, yeah, the way the system was supposed to be working of monitor is supposed to be read-only for all channels.So it's read-only for both of those channels.Draft potentially drafts a reply.That's really most relevant for our email channels or other customer channels or the internal Tabristein things in Slack.Alfred is the only one that can actually send any replies.But we were supposed to have had specific channels that are the only channels Alfred can send any replies.Nothing email.Only a couple of specific things in Slack, which actually was not set up yet, but today only in Slack.And then WhatsApp only to me.And so there was something that was not set up properly in that, in theory, Alfred knew that.That there was only supposed to be replies coming to me, yet messages were still getting sent to my wife.So we did have to do essentially that same thing of, okay, Claude, here, look at the system.Codex, look at the system.Let's review everything because this is not working.Figure out why.Propose a plan.Let's compare the plans.I like this.Now let's fix it.I'm thinking one thing that we both are facing with OpenClaw, and it's good that it's open source, so those bugs are getting found.The core part of our architecture is essentially not fully within our control.Yes.Yeah.Very reliant on the rest of the community, essentially, that is continuing to also have eyes on, hands on, seeing what still needs to be improved and helping make it happen.Well, so let's add a little bit more context into why there's having this conversation and then kind of where I think we're going next.So we both shared our open call systems and a little bit of what we've talked about here.At a local AI, so Innovation Council open call community night just last week, it was that sharing and that community that was present in the room that, to me, prompted, we should continue doing this.And maybe it doesn't have to always be theseevents in person, maybe we can have more one-on-one so we can dig into a little bit more of the details.So I did.I went back and forth with Alfred chatting about this idea.And I happen to I'm aMicrobiology was my degree in undergrad, and I really like biology.Before that, as a kid, I had thought I was going to be a marine biologist.I kept staying in science area, but so I happen to know that a RISC was a group of lobsters, and an open clause icon logo is a lobster.It felt like a great opportunity that RISC is lobsters.RISC is also inherently a theme that we heard acrossAll 5 of the presentations and that you see come up in the communities online where people were talking about it.Risk needs to be in the name of this podcast conversation.He drafted the landing page for the podcast andI had Claude Code wire it up and plug it directly into the GitHub, and the site went live.So that got set up, and then it was, all right, let's start having some conversations.You were the first volunteer.We'll see how this goes.My thought is that I will get the videos back soon.I will take the transcript of it and hand it over to Alfred to write some podcast notes.I don't know how this works or how other people do this.I intentionally didn't go look myself yet.I kind of wanted to see what the AI would come up with for taking our conversation and figuring out where it suggests that we edit it.I'll hand the raw filesAlfred's notes over to my friend that's going to help edit it and see what he thinks and we'll probably what we do for episode one is not what I think we'll end up doing for episode two and beyond we'll refine and tweak from thereVery cool.But I'm excited that you had this idea.It definitely felt after the night on March 11th that the conversation deserved a little bit more.And I definitely appreciate you sharing your story.And my aha moment was how like how differently you can integrate the whole open claw and ecosystem even based on our stories.Yeah, I think in many ways, I think our 2 feel like there's the most commonalities compared to the other 3 that were in the room.But there's still enough differences in those 2, let alone the other 3 use cases, which we've not gotten to yet.And hopefully I get those guys to also come in and share.And I know just from some of the other conversations I was having with people in attendance, there were anumber of other people already building some of them further along in many ways to me so i'm excited to keep having conversations learning building it just feels like a really cool time hundred percent yeah thanks for coming to new york thanks for having me steven this has been a lot of fun